Any tool has its limits. Scroll down on his post for a bit of a conversation in the comments.
I'd say that any committee decisions are difficult to predict, especially if the committee is not stable in membership, nor has a track record of similar decisions to quantify.
Juries might seem fit this description, but I think they are an exception, because there are a lot of legal (e.g. voir dire rules), judicial (charges to the jury from the judge), and cultural (Law & Order television) that would actually stabilize the process of selecting juries.
Image link here.